协同学习差分进化算法及在双转向机构优化中的应用

Cooperative Learning Differential Evolution Algorithm and Its Application in Optimization of Double Steering Mechanism

根据齐次坐标变换法推导了双转向机构转向分析数学模型,然后采用差分进化(DE)算法求解该模型。针对基本DE算法可能出现早熟或收敛速度慢的问题,提出一种基于协同学习机制的差分进化(CLDE)算法。该算法采用两个子种群,每个子种群采用不同的变异策略,利用局部极值判断机制确定早熟收敛种群;针对早熟收敛种群,利用精英种群映射策略向精英种群进行映射学习,实现子种群间的信息交流;若不存在精英种群,则在自身种群内采用自适应高斯扰动策略实现自我调整。函数测试结果表明,CLDE优化精度更高、速度更快、稳定性更好。将该算法用于机构优化问题,结果表明,与基本DE算法、随机变异差分进化算法(RMDE)、多种群自适应差分进化算法(ADEMP)相比,CLDE的最优适应度值分别降低13. 83%、8. 33%和6. 25%,且表现出了较好的稳定性和收敛性。

According to the homogeneous coordinate transformation method,the mathematical model of the steering analysis of the double steering mechanism was derived,and then the differential evolution( DE) algorithm was applied to solve the model. Aiming at the problem that the basic DE algorithm may appear premature or slow convergence,a differential evolution algorithm based on cooperative learning( CLDE) mechanism was presented.Two populations were used in the algorithm,and each population adopted different mutation strategies,which used a local extremum judgment mechanism to determine premature convergence population;For the premature convergence population,a elite population mapping strategy was used to map learning to the elite populations to realize information communication between the populations;If there was no elite population,a adaptive Gaussian perturbation strategy was used to achieve self-adjustment within its own population. The function test results show that the CLDE has the characteristics of high optimization precision,quick convergence and good stability. The algorithm was applied to a problem of mechanism optimization,compared with the basic DE algorithm,random mutation differential evolution( RMDE) algorithm and adaptive multi-population differential evolution( ADEMP) algorithm,the results show that the optimal fitness value of CLDE decreases by 13. 83%,8. 33% and 6. 25% respectively,and which shows good stability and convergence.